The basic modeling tool considered in this chapter remains generalized linear models. The chapter begins with the subject of data layout for a situation involving longitudinal data. The iterative routine for estimation of regression coefficients also provides iterative estimation of correlation with methodology that utilizes and is consistent with the structure that is chosen by the user. The generalized estimating equation methodology provides a consistent estimator of β, even if the correlation structure is incorrect because the procedures use consistent estimates of the variance-covariance structure. The empirical or robust estimator and the model-based estimator give similar results, which implies that the correlation structure assumed is a good practical choice. The chapter discusses the examples regarding these two types of estimators of standard errors. The foundation for the use of generalized estimating equations for nonnormal responses is similar to the one developed for normally distributed responses. Controlled Vocabulary Terms autocorrelation; estimating equations; non-linear iterative partial least squares

Free Access Appendix A.4: The Relationship Between Maximum Likelihood Estimation of the Logistic Regression Model and Weighted Least Squares Raymond H. Myers, Raymond H. Myers Virginia Polytechnic Institute and State University, Blacksburg, VirginiaSearch for more papers by this authorDouglas C. Montgomery, Douglas C. Montgomery Arizona State University, Tempe, ArizonaSearch for more papers by this authorG. Geoffrey Vining, G. Geoffrey Vining Virginia Polytechnic Institute and State University, Blacksburg, VirginiaSearch for more papers by this authorTimothy J. Robinson, Timothy J. Robinson University of Wyoming, Laramie, WyomingSearch for more papers by this author Book Author(s):Raymond H. Myers, Raymond H. Myers Virginia Polytechnic Institute and State University, Blacksburg, VirginiaSearch for more papers by this authorDouglas C. Montgomery, Douglas C. Montgomery Arizona State University, Tempe, ArizonaSearch for more papers by this authorG. Geoffrey Vining, G. Geoffrey Vining Virginia Polytechnic Institute and State University, Blacksburg, VirginiaSearch for more papers by this authorTimothy J. Robinson, Timothy J. Robinson University of Wyoming, Laramie, WyomingSearch for more papers by this author First published: 03 March 2010 https://doi.org/10.1002/9780470556986.app4Book Series:Wiley Series in Probability and Statistics AboutPDFPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShareShare a linkShare onFacebookTwitterLinked InRedditWechat Generalized Linear Models: With Applications in Engineering and the Sciences, Second Edition RelatedInformation

Background on Basic Test StatisticsWe indicate that Y is a random variable that follows a normal distribution with mean μ and variance σ 2 by Υ~Ν(μ, σ 2 ) Central Distributions1. Let Y u Y2,•., Y n be independent normally distributed random variables with E(Y¡) = jU/, and Var(T¿) = σ,• 2 .Let a^ a 2 ,..., a n be known constants.If we define the linear combination of the 7/s by U~N [Y^a^Y^aWThe key point is that linear combinations of normally distributed random variables also follow normal distributions. If Υ~Ν(μ, σ\ thenwhere Z is the standard normal random variable.

The levels used in a study for random effects represent a random sample from a much larger population of possible levels. Many studies involve mixed effects models where some regressors are fixed effects and some are random effects. Mixed effects models are useful for a wide array of study types. This chapter discusses the extension of linear regression models to linear mixed effect models and generalized linear models (GLMs) to generalized linear mixed models (GLMMs). It outlines a Bayesian approach to generalized linear mixed models. Some of the advantages afforded by this approach are emphasized. The chapter considers a Bayesian approach to modeling exponential family data with random effects. While the GLMM, hierarchical generalized linear models (HGLMs), and GEE approaches tend to focus primarily on inference for the population mean, the Bayesian approach offers a great deal of flexibility in terms of the types of distribution characteristics. Controlled Vocabulary Terms Bayesian inference; generalized linear mixed model

Applications of regression are numerous and occur in almost every applied field including engineering and the chemical/physical sciences, life and biological sciences, the social sciences, management and economics. A very important type of regression model is the linear regression model. This chapter summarizes the techniques for estimating the parameters in multiple regression models. It presents the standard methods for testing hypotheses and constructing confidence intervals for these models, as well as methods for checking model adequacy and quality of fit. The chapter describes two important parameter estimation techniques for the linear regression model such as the method of least squares and the method of maximum likelihood. It also describes the important role of the normal distribution in linear regression. The chapter ends with a discussion of designing experiments. Controlled Vocabulary Terms linear regression; maximum likelihood estimation; multiple regression

Taguchi's Parameter Design: A Panel Discussion Author(s): Vijayan N. Nair, Bovas Abraham, Jock MacKay, John A. Nelder, George Box, Madhav S. Phadke, Raghu N. Kacker, Jerome Sacks, William J. Welch, Thomas J. Lorenzen, Anne C. Shoemaker, Kwok L. Tsui, James M. Lucas, Shin Taguchi, Raymond H. Myers, G. Geoffrey Vining, C. F. Jeff Wu Source: Technometrics, Vol. 34, No. 2 (May, 1992), pp. 127-161 Published by: American Statistical Association and American Society for Quality Stable URL: http://www.jstor.org/stable/1269231 Accessed: 16/06/2010 15:53

This paper discusses the implementation of the robustness criterion of Box and Watson (1962) for first order linear models. Necessary and sufficient conditions are given for the existence of a design which satisfies this criterion, and a method is described for the construction of such a design for a given number of exper-imental runs. Some difficulties associated with the practical application of the Box-Watson criterion are also discussed.